Math 263: Infinite-Dimensional Lie Algebras

23 Pages·2012·0.275 MB·English

by  Daniel Bump

Math 263: Infinite-Dimensional Lie Algebras
Our text will be Kac, Infinite-Dimensional Lie Algebras (third edition) though I will
also rely on Kac and Raina, Bombay Lectures on Highest Weight Representations of Infinite-
Dimensional Lie Algebras. I will assume you have access to the first text but not the second.
Unfortunately Kac’ book is long and dense. These notes are an attempt to make a speedy
entry. Most relevant for the first part of these notes are Chapters’s 7 and 14 of Kac’s book.
1 Examples of Lie Algebras
Given an associative algebra A, we have two Lie algebras, Lie(A) and Der(A).
Lie(A)=A as a set, with bracket operation [x,y]=xy yx.
• −
Der(A) is the set of derivations of A. It is a subalgebra of Lie(End(A)). For this it
•
is not necessary that A be associative.
If X is a vector field on a manifold M, then X may be be identified with the map
C∞(M) C∞(M)whichisdifferentiationinthedirectionX. BytheLeibnitzrule,X(fg)=
(cid:17)
X(f) g + fX(g). So X is a derivation of C∞(M) and in fact every derivation of A =
C∞(M) arises from a vector field. Thus the space vector fields on M may be identified
with Der(A). It is an infinite-dimensional Lie algebra.
If G is a Lie group, then the vector space g of left invariant vector fields is isomorphic
to the tangent space T (G) of G at the identity. This is obviously closed under the bracket
1
operation. This is the Lie algebra of G.
Given a vector field X on a manifold M, there is a family of integral curves. That is, (at
least if M is compact) there should be defined a map φ:M ( ε,ε) M such that φ(m,
× − (cid:17)
0)=m and
d
Xf(m)= f(φ(m,t))
dt |t=0
and more generally
d
Xf(φ(u,m))= f(φ(t+u,m)) .
dt |t=0
If G is a Lie group, and X g, the Lie algebra, then X is a left-invariant vector field, and
∈
it may be deduced that the integral curve has the form
φ(g,t)=g exp(tX)
·
where exp:g G is a map, the exponential map. We also denote exp(X)=eX. So
(cid:17)
d
Xf(g)= f(g etX) .
dt · |t=0
If π:G GL(V) is a representation, then we obtain a representation of g by
(cid:17)
d
dπ(X)v= π(etX)v .
dt |t=0
1
2 Section 3
We have
dπ([X,Y])=dπ(X) dπ(Y) dπ(Y) dπ(X).
◦ − ◦
2 Attitude
For Lie groups, statements about group theory may be translated into statements about
Lie algebras. For example, the classification of irreducible representations of a semisimple
complex or compact Lie group is (at least in the simply-connected case) the same as the clas-
sification of the irreducible representations of its Lie algebra. (Representations of a complex
Lie group are assumed analytic, and representations of a compact Lie group are assumed
continuous. Representations of a complex Lie algebra are assumed C-linear, and represen-
tationsions of a real Lie algebra are assumed R-linear.
Thus to a first approximation, the finite-dimensional theories of Lie groups and Lie
algebras are nearly equivalent. The theory of Lie groups is superior for supplying intuition,
but the theory of Lie algebras is simpler since the operation is linear.
In the infinite-dimensional case, the fact that the theory of Lie algebras is simpler is
decisive. Still, we may use the theory of groups to supply intuition, so we may preface a
discussion with some optional group theory.
For example, let us motivate the definition of an invariant bilinear form with the group-
theoretic definition. We assume that we have an action of a Lie algebra on a vector space
V, that is, a map π:g End(V) such that π([X,Y])=π(X) π(Y) π(Y) π(X). Then
(cid:17) ◦ − ◦
by an invariant bilinear form we mean a bilinear map B:V V C such that
× (cid:17)
B(X v,w)+B(v,X w)=0. (1)
· ·
Here X v means π(X)v. Why is this the right definition? Begin with the well-motivated
·
and obviously correct notion for a Lie group:
B(g v,g w)=B(v,w).
· ·
Take g=etX and differentiate with respect to t:
d
0= B(etX v,etX w).
dt · ·
By the chain rule, if F(t,u) is a smooth function of two variables then
d ∂ ∂
F(t,t)= F(t,u) + F(t,u)
dt ∂t |u=t ∂u |u=t
and applying this to F(t,u)=B(etX v,euX w) we obtain (1).
· ·
3 Central Extensions and Derivations
As with groups, we may consider extensions of Lie algebras, that is, equivalence classes of
short exact sequences
0 k g q 0.
(cid:17) (cid:17) (cid:17) (cid:17)
Central Extensions and Derivations 3
Given an extension, there are two cases where there is associated an action of q on k. This
means a homomorphism from q to the Lie algebra Der(k) of derivations of k. The first is the
case where k is abelian. The second is where the extension is split. In both these cases, the
Lie algebra g may be recovered from k, q and some additional data.
If k is abelian, then k is a q-module, and the extra data needed to describe g is a 2-
•
cocycle. The special case where the q-module structure is a 2-cocycle. For simplicity
we will only discuss (below) the case where the q-module structure on k is trivial, or
equivalently that the image of k in g is central.
If the extension is split,then the only data needed to describe g is the homomorphism
•
q Der(k).
(cid:17)
As an important special case, we are interested in increasing the dimension of a Lie algebra
by one. That is, we want to consider extensions:
0 C c g˜ g 0
(cid:17) · (cid:17) (cid:17) (cid:17)
and
0 g′ g C d 0
(cid:17) (cid:17) (cid:17) · (cid:17)
where C c and C d are one-dimensional abelian Lie algebras. In the first case we may
· ·
assume that c is central, that is, [c,X]=0 for all X g. In the second case, we assume only
∈
that g′ is an ideal, that is, [g,g′] g′.
⊂
3.1 Central extensions
Let a be an abelian Lie algebra. We’re mainly interested in the case where a is one-dimen-
sional. We wish to classify central extensions
p
0 a g˜ g 0.
(cid:17) (cid:17) 1 (cid:17)
We choose a section s: g g˜, which is a linear map (not a homomorphism) such that
(cid:17)
p s=1 . Then if X,Y g consider
g
◦ ∈
φ(X,Y)=s([X,Y]) [s(X),s(Y)].
−
Clearly φ(X,Y) is in the kernel of p, which we may identify with a. We observe that φ is
skew-symmetric and satisfies the cocycle identity
φ([X,Y],Z)+φ([Y ,Z],X)+φ([Z,X],Y)=0.
Proof. We have
φ([X,Y],Z)=s([X,Y],Z) [s([X,Y]),s(Z)].
−
Since s([X,Y]) [s(X),s(Y)]=φ(X,Y) a, we may rewrite this
− ∈
φ([X,Y],Z)=s([X,Y],Z) [[s(X),s(Y)],s(Z)].
−
The cocycle identity now follows from the Jacobi identities in g and g˜. (cid:3)
A skew-symmetric map that satisfies the cocycle identity is called a 2-cycle. Let Z (g,a)
2
be the space of 2-cocycles.
4 Section 4
We see that every extension produces a 2-cocycle, and conversely given a 2-cocycle we
may reconstruct g˜ as follows. Let g˜=g a with the bracket operation
⊕
JX +δ,Y +εK=[X,Y]+δ+ε+φ(X,Y)
Then the 2-cocycle condition implies the Jacobi identity.
We have some freedom of choice in the section s. We could vary it by an arbitrary linear
map f:g a. This changes the cocycle by the map
(cid:17)
ψ(X,Y)=f([X,Y]).
A cocycle of this form is called a coboundary. If B (g,a) is the space of coboundaries then
2
we see that central extensions are classified by H (g,a)=Z (g,a)/B (g,a).
2 2 2
3.2 Adjoining a derivation
A derivation is a local analog of an automorphism. Thus, let A be an algebra, whether
associative or not. A map D:A A such that D(xy)=D(x) y +xD(y) is a derivation.
(cid:17)
Exponentiating it:
1
eD=1+D+ D2+
2 (cid:10)
then satisfies eD(xy)=eD(x)eD(y).
1
(x+D(x)+ D2(x)+ .)
2 (cid:9)
If g is a Lie algebra, then ad:g End(g) is the adjoint representation, defined by
(cid:17)
ad(X)Y =[X,Y].
It follows from the Jacobi identity that ad(X) is a derivation. The map ad is a Lie algebra
homomorphism, that is, ad([X,Y])=ad(X)ad(Y) ad(Y)ad(X). This also follows from the
−
Jacobi identity.
Suppose that k is an ideal of g of codimension 1, and let d g k. Then ad(d) induces
∈ −
a derivation of k. Conversely, suppose that g is given, together with a derivation D. Then
we may construct a Lie algebra g that is the direct sum of g and a one-dimensional space
C d, where the Lie bracket is extended to k by requiring that [D,X]= [X,D]=d(X). The
· −
Jacobi identity follows from the fact that D is a derivation.
More generally, given a pair k, q of Lie algebras and a homomorphism α:q Der(k) we
(cid:17)
may construct a Lie algebra g which contains copies of both q and k. The subalgebra k is an
ideal, and if X q and Y k, then α(Y)X=[Y ,X].
∈ ∈
4 Virasoro Algebra
Let us consider the Lie algebra w of (Laurent) polynomial vector fields on the circle T=
t C× t =1 , or equivalently, polynomial vector fields on C×. This is the Lie algebra of
{ ∈ k | }
the (infinite-dimensional) group of diffeomorphisms of the circle. It has a basis
d
d =t1−i .
i dt
Loop algebras 5
These satisfy the commutation relation
[d ,d ]=(i j)d .
i j i+j
−
Let N be an integer.
Proposition 1. Let
iN if i= j
φN(di,dj)= 0 otherw−ise,
(cid:26)
extended by linearity to a map w w C. If N =1 or 3 then φ is a 2-cocycle. (We are
N
× (cid:17)
regarding C as a one-dimensional abelian Lie algebra.
Proof. If Nis odd then φ is skew-symmetric, and we need
N
φ ([d ,d ],d )+φ ([d ,d],d )+φ ([d ,d ],d)=0.
N i j k N k i j N j k i
The left-hand side is trivially zero unless i+j+k=0, so we assume this. Then the left-hand
side equals
(i j)kN+(j k)iN+(k i)jN=(i j)( i j)N+(i+2j)iN (j+2i)jN.
− − − − − − −
This is zero if i+j+k=0 and N =1 or 3. (cid:3)
It may be seen that φ and φ span Z (w,C), but φ is a coboundary. The space H2(w,
1 3 2 1
C)isone-dimensional. ThereisauniqueisomorphismclassofLiealgebrasthatarenontrivial
central extensions of w by C. (Each element of H2(w,C) determines a unique equivalence
class of extensions, but two inequivalent extensions may be isomorphic as Lie algebras.)
The choice of extension is traditionally to take the cocycle 1 (φ φ ). Thus we obtain
12 3− 1
the Virosoro algebra with basis d ,c, where c is central, and
i
(i j)d + 1 (i3 i) if i= j,
[di,dj]= − i+j 12 − −
( (i j)di+j otherwise.
−
5 Loop algebras
In this section we will describe affine Kac-Moody Lie algebras following Chapter 7 of Kac,
Infinite-dimensional Lie algebras. This illustrates both methods of enlarging a Lie algebra:
first by making a central extension, then adjoining a derivation.
If g is a Lie algebra and A is a commutative associative algebra, then A g is a Lie
⊗
algebra with the bracket
[a X,b Y]=ab [X,Y].
⊗ ⊗ ⊗
Let g be a complex Lie algebra that admits an ad-invariant symmetric bilinear form,
which we will denote ( ). Recalling that ad:g End(g) is the map ad(X)Y =[X,Y], this
| (cid:17)
means
([X,Y] Z)= (Y [X,Z]).
| − |
6 Section 5
Since the form is symmetric, this implies
([X,Y] Z)=([Y ,Z] X). (2)
| |
For example, let G be a compact Lie group. Then G acts on itself by conjugation,
0 0
inducing a representation Ad of G on left-invariant vector fields, that is, on its Lie algebra
0
g . The representation of g on itself induced by Ad is ad. Since Ad: G GL(g ) is a
0 0 0 0
(cid:17)
representation of a compact Lie group on a finite-dimensional real vector space, it admits
an invariant bilinear form, and by the argument in Section 2, it follows that this gives an
invariant positive definite symetric bilinear form ong . Since g is actually a real Lie algebra,
0 0
we then pass to the complexified Lie algebra g=C g to obtain a complex Lie algebra.
0
⊗
Let =C[t,t−1] be the commutative algebra of Laurent polynomials. Then (g)= g
L L L⊗
is an infinite-dimensional Lie algebra. We will show how to obtain a central extension g′ of
it. The residue map Res: C is defined by
L(cid:17)
Res c tn =c .
n −1
(cid:18) n (cid:19)
We will also define X
dP
φ(P,Q)=Res(P′Q), P′= .
dt
We have
Res(P′)=0
for any Laurent polynomial. Thus (using ′to denote the derivative) since(PQ)′=P′Q+PQ′
and (PQR)′=P′QR+PQ′R+PQR′ we obtain
φ(P,Q)= φ(Q,P)
−
and
φ(PQ,R)+φ(QR,P)+φ(RP,Q)=0. (3)
These relations are similar to the conditions that must be satisfied by a 2-cocycle. So we
define ψ: (g) (g) C by
L ×L (cid:17)
ψ(a X,b Y)=(X Y)φ(a,b).
⊗ ⊗ |
This bilinear form is clearly skew-symmetric. Here is the proof of the cocycle relation:
ψ([a X,b Y],c Z)=([X,Y] Z)φ(ab,c).
⊗ ⊗ ⊗ |
Using (2) and (3) we have
ψ([a X,b Y],c Z)+ψ([b Y ,c Z],a X)+ψ([c Z,a X],b Y) =
⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗
([X,Y] Z)[φ(ab,c)+φ(bc,a)+ψ(ca,b)] = 0.
|
Let g′ be the central extension of (g) determined by this 2-cocycle. We may further
L
enlarge g′ by adjoining a derivation: use the derivation tdt of (acting by zero on the
L
adjoined one-dimensional central element. Let gˆ be the resulting Lie algebra.
Projective Representations and Central Extensions 7
If g is a simple complex Lie algebra, then gˆ is an affine Kac-Moody Lie algebra.
6 Projective Representations and Central Extensions
It is well-known that there is a relationship between projective representations of a group
and central extensions. We briefly recall this before considering the analogous situation for
Lie algebras. Suppose that π:G GL(V) is a projective representation of a group G; that
→
is, a map that induces a homomorphism G Aut(P1(V)), the group of automorphisms of
→
the corresponding projective space. Concretely, this means that
π(g)π(h)=φ(g,h) π(gh)
for some map φ:G G C. Now the map φ satisfies the cocycle condition
× →
φ(g g ,g ) φ(g ,g )=φ(g ,g g )φ(g ,g ),
1 2 3 1 2 1 2 3 2 3
which means that we may construct a central extension of G. Let G˜ be the group which as
a set is G C× with the multiplication
×
(g,ǫ)(h,δ)=(gh,ǫδφ(g,h)).
The cocycle condition implies that this group law is associative. we may define a represen-
tation π˜:G˜ GL(V) by
→
π˜(g,ǫ)v=ǫπ(g)v,
and this is a true representation. Therefore every projective representation of G may be
lifted to a true representation of a covering group, that is, a central extension determined
by a cocycle describing the projective representation.
Now let us consider the corresponding construction for a Lie algebra representation. A
representation of a Lie algebra g is a linear map π:g End(V) such that
→
π([X,Y])v=π(X)π(Y)v π(Y)π(X)v, X,Y g.
− ∈
Thecorrespondingnotionofa projective representation requiresamap φ:g g Csuchthat
× →
π([X,Y])v=π(X)π(Y)v π(Y)π(X)v+φ(X,Y)v, X,Y g. (4)
− ∈
Lemma 2. In (4), φ is a 2-cocycle.
Proof. It is straightforward to check that φ(X,Y)= φ(Y ,X). We have
−
π([[X,Y],Z])= π(X)π(Y)π(Z) π(Y)π(X)π(Z) π(Z)π(X)π(Y)
− −
+ π(Z)π(Y)π(X)+φ([X,Y],Z) I .
V
Summing cyclicly and using the Jacobi relation gives
φ([X,Y],Z)+φ([Y ,Z],X)+φ([Z,X],Y)=0. (cid:3)
8 Section 7
Remark 3. As a special case, let π: g End(V) be a true representation. We obtain
→
a projective representation by perturbing it. Thus, let λ: g C be an arbitrary linear
→
functional, and define π′(X) = π(X) λ(X). Then π′ is a projective representation with
−
φ(X,Y)=λ([X,Y]). This cocycle is a coboundary.
7 Fermions
1
The Dirac equation describes the electron or other particles of spin . Such a particle is a
2
fermion, meaning that the wave equation is skew-symmetric, changing sign when particles
are interchanged. Therefore no two particles can occupy the same state. (This is the Pauli
exclusion principle.) In the simplest case, electrons are solutions to the Dirac equation
characterized by energy (which is quantized) but no other quantum number. There is thus
one solution with energy k for each integer k. Thus if v is a vector in a suitable Hilbert
k
space V representing such a solution, a solution representing the superposition of several
electrons might be represented by an element v v of the exterior algebra on V.
k1∧ k2∧(cid:10)
A difficulty with the Dirac equation is the existence of solutions with negative energy.
Particleswithnegativeenergyareunphysicalandyet theyarepredictedbythetheory. Dirac
postulated that the negative energy solutions are fully populated. Thus the ground state is
the “vacuum vector”
0 =v v v
0 −1 −2
| i ∧ ∧ ∧(cid:10)
If one electron is promoted to a state with higher energy, we might obtain a state (for
example)
v v v v v
3 0 −1 −3 −4
∧ ∧ ∧ ∧ (cid:10)
The v then appears as an electron with energy 3. The missing particle v represents the
3 −2
absence of a usually present negative energy particle, that is, a positive energy particle
whose charge is the opposite of the charge of the electron. This energetic state represents
an electron-positron pair. This scheme avoids the physical appearance of negative energy
solutions.
We are led to consider the space of semi-infinite wedges, spanned by elements of the form
v v
im∧ im−1∧(cid:10)
where (due to the skew-symmetry of the exterior power) we may assume that
i >i >i >.
m m−1 m−2
Wewillcallsuchavectorasemi-infinite monomial. Theintegermischosensothat i +k=
−k
0 for sufficiently large k. The integer m is the excess of electrons over positrons, so m is
called the charge of the monomial. Let F(m) be the span of all monomials of charge m, and
let F be the direct sum of the F(m). The space F is called the Fermionic Fock space.
We will denote
m =v v F(m). (5)
m m−1
| i ∧ ∧(cid:10) ∈
This is the vacuum vector in F(m).
Fermions 9
Let gl be the Lie algebra spanned by elements E with i,j Z. This is an associative
∞ ij
∈
ring (without unit) having multiplication
E E =δ E , δ=Kronecker delta,
ij kl jk il
and as usual the associative ring becomes a Lie algebra with bracket [X,Y]=XY YX.
−
Elements may be thought of as infinite matrices as follows: we will represent an element
a E gl , where all but finitely many a as a matrix (a ) indexed by i,j Z.
ij ij ∞ ij ij
∈ ∈
The Lie algebra gl acts on the v by E v = δ v. Then there is an action on the
∞ k ij k jk i
P
seminfinite monomials by
X(v v )= v v X(v ) v . (6)
im∧ im−1∧(cid:10) im∧(cid:10) ∧ ik+1∧ ik ∧ ik−1∧(cid:10)
k
X
In accordance with the Leibnitz rule, we have applied the matrix X to a single v , then
k
summed over all k.
The Lie algebra gl has an enlargement which we will denote a¯. This is again obtained
∞
from an associative algebra, namely we now consider (a ) where now we allow an infinite
ij
number of nonzero entries, but we require that there exists a bound N such that a = 0
ij
when i j >N. This is sufficient to imply that in the definition of matrix multiplication,
| − |
(a ) (b )=(c ) where c = a b , the summation over k is finite.
ij · ij ij ij k ik kj
The action of gl on semi-infinite monomials does not extend to a¯. However it almost
∞
P
does so extend. Applying (6) formally gives
a v v v v . (7)
ij im∧(cid:10) ∧ ik−1∧ i∧ ik+1∧(cid:10)
 
n i,j,k
X X
|n|6N i−j=n
 ik=j 
 
 
The inner sum is finite for all n except n=0. The reason is that if n 0 then there will only
(cid:2)
be finitely many i such that replacing v by v will not cause a repeated index, and
k ik ik+n
v v v v is treated as zero unless the indices are all distinct.
im∧(cid:10) ∧ ik−1∧ ik+n∧ ik+1∧(cid:10)
However the n=0 the inner sum is
a v v v v
ik,ik im∧(cid:10) ∧ ik+1∧ ik∧ ik−1∧(cid:10)
which is problematic. Xik
We may address this problem by making use of Remark 3. We will perturb the repre-
sentation π:gl End(F) defined by the above action (so Xv=π(X)v) by introducing a
∞
(cid:17)
linear functional λ on g=g, then let
π′(X)v=π(X) λ(X)I .
F
−
Thus π′ is only a projective representation, but if λ is chosen correctly it has the advantage
that it extends to a¯.
We consider the linear functional λ:gl C defined on basis vectors by
∞
(cid:17)
1 if i=j<0,
λ(E )=
ij
0 otherwise.
(cid:26)
10 Section 8
In this case, the term n=0 in (7) becomes
a v v v v
ik,ik im∧(cid:10) ∧ ik+1∧ ik∧ ik−1∧(cid:10)
iXk>0
which is a finite sum even when (a ) a¯.
ij
∈
Now the cocycle φ(X,Y)=λ([X,Y]) is a coboundary on gl . It is given by
∞
1 if i<0,j>0
φ(E ,E )= 1 if i>0,j<0,
ij ji
 −
0 otherwise

while φ(Eij,Ekl)=0 unless j=k and i=l. Let a be the corresponding central extension of
a¯. Then we obtain a representation of a on F.
8 The Heisenberg Lie Algebra and the Stone-von Neu-
mann Theorem
Let H be the Heisenberg Lie algebra, which has a basis p , q (i = 1, 2, 3, ) and c. The
∞ i i
(cid:10)
vector c is central, while
[p ,q]=c. (8)
i i
These elements have nicknames coming from quantum mechanics: q are “position,” p
i i
are“momentum.”Thisisbecauseifwehaveindependentparticles,withcorrespondingspace-
like variables x , x , x , the position and momentum operators on the Schwartz space
1 2 3
(cid:10)
(Rn) by
S
∂
(q f)(x ,x , )=xf, p = i~
i 1 2 (cid:10) i i − ∂x
i
give a representation of H with c action by i~. Here ~ is Planck’s constant. There is a
∞
−
certain symmetry since the Schwartz space is invariant under the Fourier transform, which
intertwines the position and momentum operators.
The representation extends to tempered distributions. To obtain a purely algebraic
analog, we may restrict the representation to the space of polynomials, which are not in
the Schwartz space, but nevertheless are tempered distributions. The constant polynomial
is annihilated by the position operators, and this corresponds to the “vacuum vector.” Inter-
preted as a physical state, it corresponds to a function with definite (zero) momentum
but completely indeterminate position. The symmetry is now lost since the Fourier trans-
form of a polynomial is a tempered distribution but not a polynomial, or even a function.
Still, the polynomial representation is an algebraic substitute for the topological one.
Abstractly, given a representation of H , we call a nonzero vector v a vacuum vector if
∞
it is annihilated by the p. If cv =λv then we say λ is the eigenvalue of v. In the physics
i
literature the vacuum vector is denoted 0 (following Dirac).
| i
Similarly, let a be any constant, and consider the representation of H on polynomials
∞
∂
with q being multiplication by x and p =λ , and λ being the eigenvalue of c. This H -
i i i ∂xi ∞
module is denoted R .
λ